Claude & Gemini: AI Agents Take Over

The era of artificial intelligence simply being a chatbot you talk to in a browser tab is officially ending, and we are entering the age of the agent that actually does the work for you. I just finished watching a fascinating breakdown by an industry pro who tested a new suite of tools that can literally control your computer’s operating system. The expert highlighted how major players like Anthropic and Google are shifting from generating text to executing complex, multi-step tasks on your behalf.

This isn’t just about writing emails anymore; it is about giving an AI permission to access your file system, read your personal photos to find information, and even negotiate contracts. The creator of the video demonstrated how these new features work in real-time, and honestly, the results are both incredibly impressive and a little bit terrifying regarding privacy and control. It feels like we just jumped forward a year in terms of utility.

🤖 The Rise of “Co-Working” Agents

The biggest headline from this analysis is the launch of Claude Co-Work, a feature that essentially turns the AI into a digital intern with access to your Mac’s hard drive. The video’s author ran a live experiment that I think perfectly illustrates the power of this technology. He granted the AI access to his notoriously messy “Downloads” folder, which contained hundreds of unorganized files, and asked it to clean everything up.

What happened next was a masterclass in autonomous task management. The tool didn’t just move files randomly; it scanned the directory, analyzed the file types, and proposed a comprehensive plan. It identified 3D model files, installers, and archives, then suggested a specific folder structure and naming convention. It even flagged duplicate files and old installers for potential deletion to save space. Once the user clicked “approve,” the AI began executing terminal commands in the background. You could watch the files vanishing and reappearing in their correct folders in real-time. This is a massive shift because it shows AI operating as a layer above your applications, managing the OS itself.

While this feature was initially locked behind a steep $100 monthly plan, the expert noted that it has already started rolling out to the standard $20 Pro plan. This democratization of “agentic” capabilities means that average users, not just enterprise developers, can now automate tedious administrative tasks. However, there is a catch: it currently only works on Mac, and you have to be comfortable with an AI executing bash commands on your machine.

🧠 Google’s “Personal Intelligence” Is Here

While Anthropic is handling your files, Google is making a play for your memories. The innovator behind this update showcased Google’s new “Personal Intelligence” update for Gemini. This feature connects the AI directly to your Google ecosystem, specifically Photos, Gmail, Drive, and YouTube, to answer questions that require context about your personal life.

The most practical example shared involved a scenario where you are at a tire shop and need to know which tires to buy, but you don’t know your car’s specs. Instead of running out to the parking lot, you can simply ask Gemini, “What tires do I need?” The AI searches your Google Photos for pictures of your car, identifies the make and model from the visual data, and provides the correct tire specifications. The video host tested this himself by asking for his license plate number. The AI successfully combed through years of old photos, found the specific image of his car, and read the plate number back to him.

This is a significant leap in multimodal utility. It transforms your photo gallery from a passive storage unit into an active database of information. However, the privacy implications are the elephant in the room. The original poster emphasized that while Gemini accesses this data to answer queries, Google claims this personal data is not used to train their foundational models. Currently, this feature is available for personal Google accounts in the US, but it hasn’t fully rolled out to Workspace business accounts yet.

⚡ Industry Drama and Ecosystem Wars

The final major insight from this deep dive covers the increasingly aggressive moves companies are taking to lock users into their ecosystems. We are seeing a mix of massive partnerships and walled gardens. On one hand, Apple and Google have officially announced that Gemini will power parts of Siri, cementing Google’s dominance on both Android and iOS devices. If you use a smartphone, you are likely going to be using Gemini, whether you know it or not.

On the flip side, we are seeing friction in the developer space. The video covered some fresh drama regarding Anthropic cracking down on how developers use their API. Many coders were using Claude’s API keys inside third-party coding environments (like Open Code) because they preferred the interface. Anthropic essentially blocked this behavior, forcing users back into their own proprietary “Claude Code” interface. The expert pointed out that this upset a lot of the developer community, who felt that if they are paying for the API usage, they should be allowed to use the interface of their choice.

Simultaneously, there is corporate intrigue at OpenAI. The video detailed a strange situation where a researcher was fired from a competitor startup for alleged unethical conduct and was rehired by OpenAI on the very same day. It highlights how cutthroat the talent war has become in Silicon Valley. Between the corporate maneuvering and the tech breakthroughs, it is clear that the major AI labs are racing to capture as much market share and control as possible before the dust settles.

🛠️ Practical Use Case: The “Clean Sweep” Workflow

Based on the demonstrations in the video, here is how you can utilize these new agentic capabilities if you have access to the Claude desktop app on a Mac:

  1. Select the Target: Open the Claude desktop app and navigate to the “Co-Work” tab. You don’t need to give it access to your whole computer; start with a single messy folder, like “Downloads” or “Desktop.”
  2. The Audit Prompt: Ask the AI to “Scan this folder and propose an organization plan based on file type and project.” Do not tell it to execute yet. Let it act as a consultant first.
  3. Review the Logic: The AI will generate a list of actions (e.g., “Create folder ‘Invoices’, move .pdf files here”). Check this list carefully to ensure it isn’t archiving active project files you need on your desktop.
  4. Execute and Watch: Once approved, the AI runs the terminal commands. You can actually watch the terminal window in the sidebar to see exactly what commands are being run, which adds a nice layer of transparency to the process.

This is just the beginning of AI acting as an operating system layer.

If you want to see the full breakdown of the Google updates and the visual demos of these tools in action, you should definitely check out the full video linked below.

Scroll to Top